Two identical functions, but only one can be optimised.

[Kirin_Guess]

Dear Statalisters,

I’m learning MATA function optimisation, and run into a problem.
In short, I have two identical functions, but MATA can only optimise one of them.
I demonstrate the problem by a simple example.

MATA works well with a function like this:

mata:
   mata clear
   void FN1(todo, b, lnf, g, H) {
      lnf = ln( binomialp(189, 59, invlogit(b) ))
   }
   S1 = optimize_init()
   optimize_init_evaluator(S1, &FN1())
   optimize_init_params(S1, (0))
   optimize(S1)
end

However, MATA doesn't work with a function like this:

mata:
   mata clear
   n = 189
   r = 59
   void FN2(todo, b, lnf, g, H) {
      lnf = ln( binomialp(n, r, invlogit(b) ))
   }
   S2 = optimize_init()
   optimize_init_evaluator(S2, &FN2())
   optimize_init_params(S2, (0))
   optimize(S2)
end

The only difference between the two functions is that n=189 and r=59 are defined outside the function 2.
I've tried to deal with this problem by using optimize_init_argument, but MATA still returns the same error message: "initial values not feasible."

mata:
   mata clear
   n = 189
   r = 59
   void FN2(todo, b, lnf, n, r, g, H) {
      lnf = ln( binomialp(n, r, invlogit(b) ))
   }
   S2 = optimize_init()
   optimize_init_evaluator(S2, &FN2())
   optimize_init_params(S2, (0))
   optimize_init_argument(S2, 1, n)
   optimize_init_argument(S2, 2, r)
   optimize(S2)
end

I'd like to know why this difference matters, and how to deal with this problem, if I really need to define something outside the function. Thank you very much.

Best,

Kirin

[Christophe Kolodziejczyk]

Dear Kirin You need to define your evaluator as Void FN2(todo, b, n, r, lnf, g, H) { } More generally lnf, g and H have to be the last arguments of your evaluator. Best Christophe

[Kirin_Guess]

Dear Christophe,
Many thanks for the correction. It works.
Best,
Kirin